Ear recognition via sparse representation over learned dictionary

Feature extraction is an indispensable step in ear recognition system. In this paper, we propose to introduce sparse representation for feature extraction. Firstly, feature vectors are obtained by applying existing dimension reduction methods, then the feature vectors are used to learn the sparse dictionary, finally the sparse coding coefficients with regard to the learned dictionary are treated as the recognition feature for ultimate ear recognition. Experimental results on the USTB ear database reveal that introducing sparse representation into the extracted global feature improves the performance of ear recognition. What´s important, sparse representation over learned dictionary from downsampling features exhibit robustness regarding to noise and partial occlusion.